PyTorch Tensor to Numpy array Conversion and Vice-Versa

Pytorch is a machine learning library that allows you to do projects based on computer vision and natural language processing. In this tutorial, I will show you how to convert PyTorch tensor to NumPy array and NumPy array to PyTorch tensor.

Convert Pytorch Tensor to Numpy Array

In this section, You will learn how to create a PyTorch tensor and then convert it to NumPy array. Let’s import torch and create a tensor using it.

import torch
tensor_arr = torch.tensor([[10,20,30],[40,50,60],[70,80,90]])
tensor_arr

The above code is using the torch.tensor() method for generating tensor. There are two ways you can convert tensor to NumPy array.

By detaching the tensor.

numpy_array= tensor_arr.cpu().detach().numpy()
numpy_array

Output

Here I am first detaching the tensor from the CPU and then using the numpy() method for NumPy conversion. The detach() creates a tensor that shares storage with a tensor that does not require grad. The above tensor created doesn’t have a gradient.

Using Simple numpy() Method.

numpy_arr2 = tensor_arr.numpy()
numpy_arr2

The above tensor I have created does not have required_gradient = True that it is false. But if you are also using the gradient then you have to use tensor.detach().numpy() method. It is because tensors that require_grad=True are recorded by PyTorch.

Below is the complete code for tensor creation and conversion.

Creation of tensor

# cpu tensor with requires_grad=True
tensor_arr2 = torch.tensor([[1.,2,3],[4,5,6],[7,8,9]],requires_grad=True)
tensor_arr2

Conversion to NumPy array

numpy_arr3 = tensor_arr2.detach().numpy()
numpy_arr3

Output

Convert Numpy Array to Pytorch Tensor

The above example was for converting PyTorch tensor to NumPy array. In this section, you will learn to convert a NumPy array to a tensor.

Let’s create a NumPy array. To create a NumPy array you have to use the numpy.array() method. Execute the following code.

numpy_array = np.array([[1,2,3],[4,5,6],[7,8,9]])
numpy_array

Conversion of NumPy array to PyTorch using from_numpy() method

There is a method in the Pytorch library for converting the NumPy array to PyTorch. It is from_numpy(). Just pass the NumPy array into it to get the tensor.

tensor_arr = torch.from_numpy(numpy_array)
tensor_arr

Output

Conversion of NumPy array to PyTorch using CPU

The above conversion is done using the CPU device. But if you want to get the tensor using GPU then you have to define the device for it. Below is the code for the conversion of the above NumPy array to tensor using the GPU.

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")# check Cuda is available or not

tensor_with_gpu=torch.from_numpy(numpy_array).to(device)
tensor_with_gpu

Output

Conversion of NumPy array to PyTorch using GPU

These are the examples for the conversion of tensor to NumPy array and vice-versa. Hope you have liked this tutorial. If you have any queries then you can contact us for more information.

Source:

Official Pytorch Documentation